search

ucascnic / CudaOT

Repository for solving discrete optimal transport problems via Cplex, Sinkhorn, FastEMD and the GPU implementation of SparseSinkhorn,
53 stars 2 forks source link

Question: which method did you use to get a mapping of images to a 2-D space as shown in your paper (M3S Method)? #2

Closed YoannRandon closed 10 months ago

YoannRandon commented 10 months ago

Hello,

I work on a project where I want to evaluate a dataset diversity using metrics like Wasserstein distance or MMD. I bump into your paper and I found that the M3S method is very interesting for my subject. I just have a point I didn't fully understand. For most of codes I've seen so far, OT algorithm require to reduce the dimension of images into a certain amount of feature (MMD) or to reduce to a single value (Wasserstein 1-D) by using models like VAE or W-GAN. My question is which method did u used to reduce your HD image in a 2-D space?

Thanks